GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
farazmust123
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
farazmust123
Model Tree
Qwen/Qwen2.5-0.5B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Draft model paired with farazmust123/qwen2.5-1.5b-trialgpt-lora for greedy speculative decoding. Same tokenizer and answer format (label-first JSON) as the target; trained with the same recipe (LoRA r=16, 3 epochs, best epoch by validation macro-F1) on the training split of ncbi/TrialGPT-Criterion-Annotations. Fine-tuning raised draft-token acceptance on screening prompts from 75.7% (base 0.5B) to 92.6% (k=4). Speculative decoding output is identical to the target's greedy output by construction.
Code: https://github.com/fsdev87/generative-ai-A2 — educational research prototype, not for clinical use.